AI for Online Schools:
Auto-Generate Courses, Quizzes, and Lesson Covers

I'm Elena, a soft skills curriculum designer for corporate online schools. In four years in EdTech, I've built 27 courses from scratch — from the first draft outline to final LMS export. Before, an 8-lesson course took me 40 hours of pure work and 2-3 calendar weeks. Now, I make the same course in 8 hours and 4 workdays — with the same quality and a much better presentation.

It's not magic, and it's not cutting corners. It's a rebuilt workflow centered around AI models in one Telegram bot. Below, I'll break down my process step-by-step: which models work at what stages, what I generate entirely with AI, where I still get hands-on, and how many credits an average 8-lesson course costs.

Step 1. Lecture Material from Bullet Points

The toughest part of any course is writing eight lecture texts, each 1500-2000 words long. That used to eat up 16 out of 40 hours. Now, Claude 4.5 Sonnet in Quantium handles 12 of those hours. It holds long context and maintains logical flow for educational content way better than GPT-5.

Here's the algorithm: The expert author and I spend an hour putting together a bullet-point plan for each lecture — 12-15 points with examples, facts, and transitions. It's not the full text; it's just the skeleton. Then, the prompt for Claude:

Claude spits out a draft in 90 seconds. The expert and I then spend 30 minutes going over it, fine-tuning the factual details. That's 35-40 minutes per lesson instead of two hours. For an 8-lesson course, that's 5 hours instead of 16.

Step 2. Lesson Covers in a Series

My in-house designer used to create the covers: 1.5 hours for 8 covers in a unified style, costing 5000-8000₽ per course on my internal billing. Now, I do them myself in 50 minutes using Gemini Image.

Here's my method: First, I generate one reference cover with a detailed style description. Once I like it, I copy the prompt template and just change the subject. For a unified style series, the template looks like this:

We only change the part in brackets: 'two people talking at a table,' 'manager and team in a meeting,' 'person looking at a results chart.' Fifty covers in a unified style in an hour? Totally doable. Learn more about working with Gemini Image in our Midjourney alternatives comparison.

Step 3. Quizzes with Explanations for Wrong Answers

A good quiz isn't just about picking the right answer; it's about picking the right one and understanding why the others are wrong. I used to spend an hour per lesson writing these: a question, 4 answers, and an explanation for each. That's 8 lessons x 1 hour = 8 hours of work.

Now, Claude 4.5 handles this entire section. Here's the prompt:

I get a ready-made quiz in 60 seconds. Checking and editing takes 10 minutes. That's 8 lessons x 15 minutes = 2 hours instead of 8. Plus, the quiz quality is often better than mine — Claude catches typical cognitive biases I'd miss.

Step 4. Script Voiceovers via TTS

Video lectures need voiceovers. I used to hire a voice actor (2000₽/min of finished audio) or record it myself and then clean up the noise. For 8 lessons at 5 minutes each, that's 40 minutes of audio, 80,000₽, or 8 hours of my time.

Now, it's ElevenLabs TTS in Quantium. I pick a native Russian voice, paste the text, and get the file in a couple of minutes. 40 minutes of audio costs about 3200 credits and 1.5 hours of my time to fine-tune intonations and pauses. That's huge savings, both in money and hours.

The main thing is to specify intonation at the start of each block: "(calm, measured)" for the lecture part, "(emphasized)" for key points. Without that, the voice sounds like a flat text-to-speech reader, and students fall asleep in 90 seconds. Learn more in the article on TTS podcasts.

Step 5. Ad Teasers via Sora

Once the course is ready, you need to sell it. I used to order teaser shoots from a video production company, which cost 50-150 thousand rubles and took 2 weeks for approvals. Now I make 3-5 ad teasers, 5-10 seconds long, using Sora 2 and test them with targeted ads.

Here's an example prompt for a team management course teaser:

For 10-20 credits, I get a shot that used to cost 30,000₽ for a single day of shooting. Then I A/B test them in the ad cabinet to see which teasers get the best CTR and CPL. Production costs drop by 20-30 times. Learn more about working with video models in the Sora 2 vs. Veo 3 comparison.

Numbers: Then and Now

Save 30 hours per course. At my 2500₽/hour rate, that's 75,000₽ for each one. Our school churns out 12 courses a year, freeing up 900,000₽. We reinvest that time into new courses and making current ones even better.

Workflow in One Bot

The biggest plus for me with Quantium isn't just one model; it's a whole assembly line. I don't jump between ChatGPT, Midjourney, ElevenLabs, and Sora. Everything's in one Telegram bot, with a single credit balance. Course stages flow one after another, keeping all the context:

• Lesson bullet points → Claude generates lecture texts
• From texts → Claude builds quizzes for each lesson
• Lesson topics → Gemini Image creates cover series
• Final scripts → ElevenLabs TTS provides voiceovers
• Course description → Sora 2 generates promo teasers
• Course description → GPT-5 writes SEO descriptions and landing page copy

These aren't just individual tools anymore; it's a production chain. And it all fits into one bot, no separate subscriptions for each service.

Course Quality: Any Drop?

The main concern I hear from colleagues: "Doesn't quality suffer?" My honest answer: It doesn't, if AI is properly integrated into the process. But it does if you try to use AI to replace your instructional designer.

Here are specific metrics for our courses over six months with AI workflow:

• Student NPS stayed at 68-72 (was 70 on average pre-AI). That's within statistical error.
• Course completion rates jumped from 41% to 47%. I chalk that up to better cover quality and clearer quizzes with explanations.
• Complaints about "AI-generated" content? Just two cases in six months. Both were my fault for not refining Claude's texts enough. Now, an expert review of the final text is a mandatory checklist item.

Quality doesn't drop if you stick to two rules. First: AI drafts, an expert verifies. Second: The course's final voice always belongs to the author, not the model. Tone, jokes, references, personal stories — you add those manually on top of the AI structure. Without that, a course becomes "encyclopedic" — technically correct, but emotionally dead. And in EdTech, emotion drives half your retention.

Here's something else I noticed: AI-explained quizzes boost completion rates. Before, students just saw "you answered incorrectly" and moved on. Now they get a detailed explanation of why their logic was wrong — it's like a mini-lesson right inside the quiz. Completion rates for the final exam jumped 6 percentage points, directly because of those AI quiz explanations.

Top Objections from Management

When I brought AI into the school, the CEO and producers had three objections. All fair points, and all resolved in practice.

"Students will be offended the course was made by AI." They weren't. The course isn't "made" by AI — it's made by an instructional designer *with* AI. It's like saying "the doctor treated with an MRI." Nobody thinks the MRI does the treating.

"The quizzes will be superficial." Turns out, it's the opposite: Claude generates quizzes on typical cognitive biases better than most instructional designers. The key is giving the model the full lecture text for context, not just a description.

"This will save hours, but eat up our AI budget." The actual cost for 10-15 courses a year with a Premium subscription? Less than a junior instructional designer's monthly salary. So we didn't "add an expense"; we replaced part of the team that just couldn't keep up.

Find out more about plans on the pricing page. Related materials: marketer case study, FLUX prompt guide.

Frequently Asked Questions

Read also

AI for SMM: 20-hour case study

12 min

AI for Bloggers

13 min

TTS Podcasts: No Narrator Voiceovers

8 min